Neural ExpeditionSubscribe
TEST RECORD · ← all field reports

FIELD REPORTS/IMAGE/ISSUE #65

Re-pose a subject from a reference image and depth map

This one is useful if you have ever wanted image generation to follow two separate instructions at once: keep the person or object from one reference image, but follow the pose or structure from another. The public Space makes that workflow easy to test before you decide whether to set up the heavier local path.

MODELRefControl FLUX.2 Klein Depth
PUBLISHEDJune 29, 2026
READ TIME2 min
TESTED BYNeural Expedition
CATEGORYIMAGE

Field notes

01What it does

RefControl FLUX.2 Klein Reference Depth is an image editing workflow built around two inputs. The reference image tells the model what identity, clothing, style, or object details to preserve. The depth map tells it what pose, structure, and composition to follow.

That makes it more targeted than a normal image prompt. For example, you can start with a character portrait, provide a second image as the pose guide, and ask for a new scene where the character keeps a recognizable look instead of being regenerated from scratch. The Space can also make a relative depth map from a normal photo, so you do not need to prepare one before the first test.

02How to try it

Start with the public RefControl FLUX.2 Klein Space. Upload a reference image in the first slot, then either upload a depth map or use the built-in depth-map helper on a normal photo. For a useful first test, pick two images with roughly similar proportions. If the depth guide is much larger, smaller, or shaped very differently from the reference subject, the result is more likely to drift.

For local use, the backing LoRA weights and ComfyUI workflow are public. The workflow runs on FLUX.2 Klein, so local reproduction is realistic for readers who already use GPU image workflows, but it is not a lightweight CPU test. The practical first step is the browser demo.

03Caveat

The workflow is most convincing when the reference and depth guide are compatible. Big scale changes, unusual proportions, or crowded scenes can still cause identity drift or awkward structure.

04What you can do with it

  • Re-pose a character while keeping clothing, face shape, or visual style closer to the original.
  • Create storyboard frames where the same subject appears in different compositions.
  • Test product or prop placement against a new structure without rebuilding the subject from scratch.
  • Use a normal photo as a quick depth guide when you do not already have a prepared depth map.
  • Compare prompt-only generation against a more controlled reference-plus-depth workflow.

Try the demo

View model page